sdjournal: SeekTail should be followed by Previous, not Next

[stevenjm]

sd_journal_next is supposed to return 0 (i.e. EOF) if called at the end of the journal, which would make this issue immediately apparent. However, there is an open bug that causes it to advance to the wrong message instead (not the last one in the journal) after a call to sd_journal_seek_tail, which causes code which follows the documented use here to produce non-obviously incorrect results (see systemd/systemd#9934).

Instead, sd_journal_previous will correctly seek to the last journal entry, with the unavoidable race condition that it may not actually be the last entry if the journal is written to between calls to sd_journal_seek_tail and sd_journal_previous.

This can be verified by comparing the output of journalctl -o json | tail -1 | jq -r .MESSAGE_ID with that of the following simple Go program

package main

import (
        "fmt"

        "github.com/coreos/go-systemd/sdjournal"
)

func main() {
        j, err := sdjournal.NewJournal()
        if err != nil {
                panic(err)
        }

        err = j.SeekTail()
        if err != nil {
                panic(err)
        }

        n, err := j.Previous()
        if err != nil {
                panic(err)
        }

        fmt.Printf("went back by %d\n", n)

        e, err := j.GetEntry()
        if err != nil {
                panic(err)
        }

        fmt.Printf("found message id %s\n", e.Fields["MESSAGE_ID"])
}

and noting that the message IDs are equal (again, with an unavoidable race condition if a message is written in between).

[pires]

LGTM

[lucab]

LGTM. Can you please rebase on current master so that newer GH actions can run on this?

[lucab]

Nevermind, I managed to manually re-sync the CI. All checks are green, merging.